Chapter 12: The Principle of Computational Equivalence

Section 8: Undecidability and Intractability

Density of difficult problems

There are arguments that in an asymptotic sense most instances chosen at random of problems like limited-size PCP or tiling will be difficult to solve. In a problem like satisfiability, however, difficult instances tend to occur only on the boundary between cases where the density of black or white squares implies that there is usually satisfaction or usually not satisfaction. If one looks at simple instances of problems (say PCP with short strings) then my experience is that many are easy to solve. But just as some fraction of cellular automata with very simple rules show immensely complex behavior, so similarly it seems that some fraction of even simple instances of many NP-complete problems also tend to be difficult to solve.

From Stephen Wolfram: A New Kind of Science [citation]